Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Evidence-based reasoning
Process by which an explicit QA system analyzes and connects the provided text excerpts as evidence to build a logical and justified response.
Chain-of-Thought (CoT)
Technique where the model generates a sequence of intermediate reasoning steps before giving the final answer, making its logic transparent and verifiable.
Fact-checking
Mechanism for validating the accuracy of the claims in an answer by cross-referencing them with a knowledge base or reliable external sources.
Explainability through counterexamples
Explanation method that provides examples of cases where the answer would be different, helping the user understand the limitations and application conditions of the reasoning.
Confidence scoring grid
Scoring system that quantifies the model's level of certainty for each segment of its response, often based on the coherence of the evidence and the reliability of the sources.
Justification tree
Hierarchical structure that represents how primary evidence supports intermediate conclusions, which in turn lead to the final answer.
Question decomposition
Process that consists of breaking down a complex question into simpler sub-questions, each being solved and justified individually before recombining into the overall answer.
Source attribution
Practice of linking each part of an answer to the specific documentary source (document, paragraph) from which it was extracted.
Abductive reasoning
Type of logical inference where the system generates the most plausible hypothesis to explain a set of provided observations or evidence.
Meta-explanation
Second-level explanation that describes why the system chose a particular reasoning strategy or a specific type of evidence to answer.
Evidence consistency
Measure of the degree of logical compatibility between the different text excerpts used as evidence to support a single answer.
Synthetic evidence generation
Ability of a system to create summaries or reformulations of evidence to make them more digestible while preserving their justificative value.
Justificative entity mapping
Technique that identifies and links key entities (people, places, dates) from the question to their occurrences in the evidence to strengthen the justification.
Semantic coverage analysis
Evaluation that measures to what extent the provided evidence covers all the semantic aspects of the question asked.
Multi-hop reasoning
Ability to chain several deduction steps, where each step is justified, to answer questions requiring inference over multiple relationships.
Justifiability score
Metric that evaluates the quality of an explanation based on its clarity, relevance, and its ability to convince a user of the answer's validity.
Contrastive explanation
Method that explains not only why an answer is correct, but also why other plausible alternative answers are incorrect.
Reasoning traceability
Property of a system that allows chronological reconstruction of each step of the decision process, from the question to the justified answer.